The European Network and Information Security Agency, ENISA, published the ARTIFICIAL…

← Alle Beiträge
IT-Sicherheit · EN

The European Network and Information Security Agency, ENISA, published the ARTIFICIAL INTELLIGENCE AND CYBERSECURITY RESEARCH identifying research issues and gaps.

According to ENISA, the results will be shared and discussed with stakeholders as proposals for future policy and funding initiatives at the level of the EU.

Identified issues:

1. Verifiability, reliability, explainability, auditability, robustness, and unbiasedness of AI.
2. Quality of datasets, ensuring both algorithmic usability and accurate representation of the problem.
3. Achieving end-to-end protection for data during transit.
4. Real-world accuracy optimization instead of relying on simulated environments.
5. Addressing computational complexity and low-latency operation for critical systems.
6. Investigating model validity, bias, and time variance changes.
7. Assessing security mechanisms through standardized frameworks and diverse malicious attempts.
8. Preserving privacy, including training data and system information confidentiality.

Identified research gaps:

1. Constructing effective AI models with limited data, moving from big data to small data environments.
2. Exploring raw data approaches to minimize or eliminate the need for feature engineering and domain expertise.
3. Incorporating change detection and adaptation mechanisms for addressing non-stationarities.
4. Periodically assessing model validity to detect and address potential biases and vulnerabilities.
5. Developing methods to remove biases and imbalances that degrade model performance.
6. Creating standardized datasets for reproducibility and reliable comparison of AI solutions.
7. Distinguishing malicious attacks from faulty states.
8. Understanding the impact of increased system scale on AI efficacy, accuracy, and computational complexity.
9. Modeling interdependent cyber-physical systems to assess vulnerability impact.
10. Establishing a standardized performance evaluation framework for comparing similar solutions.
11. Providing context awareness in machine learning for enhanced resiliency.
12. Involving humans in the loop through real-world scenario-based training for practitioners.

#cybersecurity #ai

Zu diesem Beitrag gibt es auf LinkedIn ein Dokument als Anhang. Auf LinkedIn ansehen →

Auf LinkedIn diskutieren